UK Intellectual Property Office sets out new guidelines for patent applications related to Artificial Intelligence
How will the new guidance impact the patentability of AI-related inventions?
The question
How will the new guidance impact the patentability of AI-related inventions?
The key takeaway
Inventions in AI are patentable the same way that inventions in software are. The IPO recognises different categories of AI inventions – Applied AI, Core AI and training AI. In all cases, the IPO will apply similar principles, ie assessing the technical contribution of the task or process the AI invention performs when it runs.
The background
UK patent law comprises legislation, such as the UK Patents Act 1977 (the Act), as well as case law. A key tenet of patent law is that inventions related only to a scientific discovery, mathematical method, computer program, or business method are not patentable.
However, technology has rapidly evolved since the Act and the IPO has recognised that many types of computer-based inventions, while rooted in mathematical methods and computer programs, ought to be patentable. It has thus used various “Signposts” (originating in case law) to determine whether software is patentable. The IPO will apply these Signposts to the substance, rather than the form, of the invention and consider whether it makes a technical contribution to the known art.
The new guidance fits into the UK’s wider “National AI Strategy” and seeks to incentivise growth in the AI industry by providing clarity for those applying for AI-related patents and strengthening the relevant IP protections.
The development
The UK IPO is approaching AI in much the same way that it previously approached software and is employing the same Signposts and “substance over form” principles. The new guidance creates two broad categories for AI inventions, “Applied AI” and “Core AI”.
Applied AI relates to inventions that either have specific use cases outside of the running of the computer (eg using AI to build an image classifier) or make the computer work better generally.
Core AI, by contrast, does not have a particular use case but relates to “an advance in the field of AI itself (eg an improved AI model, algorithm, or mathematical method)”. For a Core AI invention to be patentable, it must contribute more than simply being a better or more well-written programme. It must itself bring about a technical effect.
The guidance also clarifies the patentability of AI-related inventions that do not fall into the categories above, such as AI training methods or data sets. Here, the guidance likens these inventions to “calibration”, which is already patentable under UK law. This has significant implications for industry, as this is one area which was lacking clarity. As large language models and text-to-image models become larger, they also become more data-hungry. The ability to patent specific datasets, or the methods used to create these data sets, will certainly be a boon to those creating AI.
Helpfully, the guidance also comes with a set of example cases, demonstrating the principle of substance over form in both “Applied” and “Core” AI contexts and affirms the patentability of many AI use-cases, such as predictive text generation.
Why is this important?
By treating the AI inventions of the 2020s in much the same way that it has treated the software inventions of the 2010s, the IPO has paved the way for industrialists and lawyers to employ many of the same standards and heuristics already in use to this new domain. The new guidance also answers some specific questions related to the patentability of various AI-related inventions.
Any practical tips?
The IPO has confirmed that, in general, patents are available for AI inventions in all fields of technology. Therefore, inventors should always explore the potential for patenting advancements in AI and ensure that patent applications are thoughtfully crafted with the existing signposts in mind.
Winter 2022